speeding up short data transfers yin zhang, lili qiu cornell university srinivasan keshav ensim...
TRANSCRIPT
![Page 1: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/1.jpg)
Speeding Up Short Data Transfers
Yin Zhang, Lili Qiu
Cornell University
Srinivasan Keshav
Ensim Corporation
NOSSDAV’00, Chapel Hill, NC, June 2000
Theory, Architectural Support, and Simulation Results
![Page 2: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/2.jpg)
2
Outline
MotivationRelated WorkTheoryArchitectural SupportSimulation ResultsConclusions and Future Work
![Page 3: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/3.jpg)
3
Motivation
Dominance of Web data transfersShort & bursty [Mah97]Small downloading time is important!
Dominance of TCPProblem: Short data transfers interact poorly with TCP !
![Page 4: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/4.jpg)
4
TCP/Reno Basics
Slow StartExponential growth in congestion window,Slow: log(n) round trips for n segments
Congestion Avoidance
Linear probing of BW
Fast RetransmissionTriggered by 3 Duplicated ACK’s
![Page 5: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/5.jpg)
5
Related Work
P-HTTP [PM94]Avoid repeated probing only for components within the SAME page.
T/TCP [Bra94]Cache connection count, RTT
TCP Control Block Interdependence [Tou97]:Cache cwnd, but large bursts cause losses
Rate Based Pacing [VH97]4K Initial Window [AFP98]Fast Start [PK98, Pad98]
Most similar to our work, but need router support to ensure TCP friendliness
![Page 6: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/6.jpg)
6
Our Approach
Directly enter Congestion AvoidanceChoose optimal initial congestion window
A Geometry Problem: Fitting a block to the service rate curve to minimize completion time
![Page 7: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/7.jpg)
7
Optimal Initial cwnd
Minimize completion time by having the transfer end at an Epoch boundary.
![Page 8: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/8.jpg)
8
Shift Optimization
Minimize initial cwnd while keeping the same integer number of RTT’s
Before optimization: cwnd = 9
After optimization: cwnd = 5
![Page 9: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/9.jpg)
9
Effect of Shift Optimization
![Page 10: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/10.jpg)
10
TCP/SPAND
Estimate network state by sharing performance information
SPAND: Shared PAssive Network Discovery [SSK97]
Directly enter Congestion Avoidance, starting with the optimal initial cwndAvoid large bursts by pacing
![Page 11: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/11.jpg)
11
Implementation Issues
Scope for sharing and aggregation24-bit heuristicnetwork-aware clustering [KW00]
Collecting performance informationNew TCP option, Windmill’s approach, …
Information aggregationSliding window average
Retrieving estimation of network stateExplicit query, active push, …
PacingLeaky bucket based pacing
![Page 12: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/12.jpg)
12
Opportunity for Sharing
MSNBC: 90% requests arrive within 5 minutes since the most recent request from the same client network (using 24-bit heuristic)
![Page 13: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/13.jpg)
13
Cost for Sharing
MSNBC: 15,000-25,000 different client networks in a 5-minute interval during peak hours (using 24-bit heuristic)
![Page 14: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/14.jpg)
14
Simulation Results
MethodologyDownload files in rounds
Performance MetricAverage completion time
TCP flavors consideredreno-ssr: Reno with slow start restartreno-nssr: Reno w/o slow start restartnewreno-ssr: NewReno with slow start restartnewreno-nssr: NewReno w/o slow start restart
![Page 15: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/15.jpg)
15
Simulation Topologies
![Page 16: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/16.jpg)
16
T1 Terrestrial WAN Link withSingle Bottleneck
![Page 17: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/17.jpg)
17
T1 Terrestrial WAN Link withMultiple Bottlenecks
![Page 18: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/18.jpg)
18
T1 Terrestrial WAN Link with Multiple Bottlenecks and Heavy
Congestion
![Page 19: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/19.jpg)
19
TCP Friendliness (I)Against reno-ssr with 50-ms
Timer
![Page 20: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/20.jpg)
20
TCP Friendliness (II)Against reno-ssr with 200-ms
Timer
![Page 21: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/21.jpg)
21
Conclusions
TCP/SPAND significantly reduces latency for short data transfers
35-65% compared to reno-ssr / newreno-ssr20-50% compared to reno-nssr / newreno-nssrEven higher for fatter pipes
TCP/SPAND is TCP-friendlyTCP/SPAND is incrementally deployable
![Page 22: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/22.jpg)
22
Future Work
Real implementation for TCP/SPANDBetter information aggregation
Exponential decay when there is not enough feedback
Understand pacing for short flows
![Page 23: Speeding Up Short Data Transfers Yin Zhang, Lili Qiu Cornell University Srinivasan Keshav Ensim Corporation NOSSDAV’00, Chapel Hill, NC, June 2000 Theory,](https://reader030.vdocument.in/reader030/viewer/2022032605/56649e7b5503460f94b7ce0e/html5/thumbnails/23.jpg)
23
Acknowledgement
Brad KarpGeoffrey M. VoelkerVenkata N. Padmanabhan